Why ChatGPT won't kill Google
But it will still have an enormous impact.
Technology hype cycles
The Gartner hype cycle is a rough outline of the stages of enthusiasm for new technology. The phases are the trigger, inflated expectations, disillusionment, enlightenment, and finally productivity. At the moment, we’re (likely) in the inflated expectations phase with regard to large language models (LLMs) like OpenAI’s ChatGPT.
The immense success of ChatGPT caught a lot of people in the tech industry off guard. From their perspective, ChatGPT’s core tech already existed (it’s predecessor, GPT3, had been around for months) - all that changed was the interface. The most caught off guard might have been OpenAI itself:
"I'll admit that I was on the side of, like, I don't know if this is going to work," OpenAI co-founder and president Greg Brockman told Fortune.
Mira Murati, OpenAI's chief technology officer, added to Fortune, "This was definitely surprising."
Brockman told the magazine that the idea to release ChatGPT for public use a few months ago was a bit of a last resort after OpenAI ran into some initial hurdles with the AI chatbot.
A big reason for the surprise is that the latest GPT models are competent at far more tasks than their creators expected. It's one thing to compose a short poem or muddle through a conversation, like other chatbots. It's another to pen a well-argued essay, decipher meaning from poetry, write a snippet of code, and cite historical events all in one sitting.
And with that, ChatGPT has captured the public's imagination - how far can this technology go? Can it really change everything? It's a fair question!
But the last few waves of technology hype have tempered my expectations a bit. In recent years, startups have trumpeted technologies like 3D printing, drones, VR, blockchain, chatbots, the metaverse, and web 3 (a.k.a. "blockchain 2"). None have (yet) lived up to their lofty expectations, though not for lack of trying.
Now, some are already suggesting that AI is the latest tech bubble. That’s a bit hard to believe at the moment, but it's not a great look when Buzzfeed's "ChatGPT for quizzes" announcement triggers a 300% stock jump1.
So while there's plenty of buzz around ChatGPT and its potential... there isn’t yet much evidence that it will completely upend everything we know and love.
But there's a second angle to the latest AI hype, a more nuanced question that's worth considering. Will ChatGPT disrupt Google's search dominance?
The death of Google?
When people were first exposed to ChatGPT, they struggled to figure out what to say to it. Given that it was a chatbot, though, they got around to asking it questions about the world. And to their surprise, it gave human-sounding answers - far better than any chatbot that had come before it.
From there, it was easy to see how large language models like ChatGPT could upend the entire way we search the web. For the past twenty-something years, we've been living in Google's world of ten blue links. But what if we stopped dealing with links entirely? What if we could ask ChatGPT, or Bing, or whoever - and get back an answer2? Would Google come out on top of that paradigm shift? And if they did, what would happen to their multi-billion dollar money printer - the ads they place above those ten blue links?
Five years ago, many people would've assumed that Google would take the lead in this new space, given its history of AI research and development. But recent events are now stacking the media narrative against Google. ChatGPT was developed by a smaller, more nimble company, OpenAI, which Google can't simply acquire. Then OpenAI loudly partnered with Microsoft, which is having something of a renaissance. And Google's response looks defensive and desperate - they've called in the founders and made an underwhelming product announcement. From CNBC:
Google employees are criticizing leadership, most notably CEO Sundar Pichai, for the way the company handled the announcement this week of its ChatGPT competitor called Bard.
Staffers took to the popular internal forum Memegen to express their thoughts on the Bard announcement, referring to it as “rushed,” “botched” and “un-Googley,” according to messages and memes viewed by CNBC.
As a result, many are wondering whether AI might finally be the Google-killer. It might happen via the partnership between OpenAI and Microsoft. Or it might come via some new upstart search engine, with AI baked in from day one.
But there's a strong argument for why Google will be fine.
First, remember the hype cycle. We're still finding our bearings when it comes to using LLMs for search. And while there are exciting signs, there are also discouraging ones. ChatGPT and Bing have a hallucination problem - you can't trust their answers to always be true. Generally speaking, your search engine straight-up lying to you is a pretty serious issue. While Google may be serving up links to the internet's misinformation, at least it isn't creating the misinformation.
Second, we're discovering some of the limits of LLMs and the chat interface. Prompt hacking is a brand new category of software exploit. It's when users "trick" ChatGPT/Bing into bypassing its internal rules, like not generating offensive content. Prompt hacking is keeping Microsoft on their toes - they've already added stricter limits to what Bing will answer, or how long a chat can last. It's not at all clear that chatting with an AI is going to be the best way (or even a good way) to search the internet. Here’s Ben Thompson on an experience interacting with Bing Chat (and its alter ego Sydney):
I’m not going to lie: having Bing say I am not a good person was an incredible experience (and for the record, I think this is another example of chatbot misinformation!). It also, to say the least, seems incredibly ill-suited to being a search engine. Microsoft (or Google) probably don’t want to be telling their users they are not a good person, and I don’t want to accidentally do a search and miss out on interacting with Sydney!
Last, default user behavior plays a critical role. Even if ChatGPT solved its hallucination and prompt hacking issues tomorrow, that's not enough to dethrone Google. Millions, if not billions, of people have the muscle memory of opening google.com or typing queries into their Chrome search bar. And that muscle memory will give Google a pretty long window to figure out its answer to ChatGPT. It doesn't have to be the first AI-powered search, as long as it can create something "good enough" before it bleeds too many users3.
The most compelling reason to doubt Google’s strength here is the fact that they seem to be playing into with Microsoft’s narrative. They’re in the dominant position, but they’re losing the plot. Cory Doctorow is also wondering why:
The really remarkable thing isn't just that Microsoft has decided that the future of search isn't links to relevant materials, but instead lengthy, florid paragraphs written by a chatbot who happens to be a habitual liar – even more remarkable is that Google agrees.
Microsoft has nothing to lose. It's spent billions on Bing, a search-engine no one voluntarily uses. Might as well try something so stupid it might just work. But why is Google, a monopolist who has a 90+% share of search worldwide, jumping off the same bridge as Microsoft?
Focusing on ChatGPT for general-purpose search, though, is missing the point.
Search is the wrong place to look
100% factual accuracy is a near-impossible goal - we can't even agree on what's "true" as humans, so why would we expect software to be better? But if you look outside of search, you can find LLMs already making a positive impact in other areas4.
Since we've seen so many "state of the art" chatbots before, there's a trap that we fall into (or at least I did). We want to see if we can "outsmart" it, so we ask it questions about things that we already know, as a test. I was 100% guilty of this - my first experiments with ChatGPT and Github Copilot left me unimpressed, because I wanted answers that I already knew.
Then I tried asking it a question where I didn't already know the answer. And something clicked. ChatGPT wasn't going to help with tasks I did every single day, which I'm an expert at. But it would be invaluable with tasks where I was stumbling in the dark. Tasks where I was struggling with a learning curve for the first time.
Never written a grant proposal? Don’t know how to do Excel formulas? Struggling to remember Java syntax? It doesn’t matter if ChatGPT hallucinates or is only 80% accurate, because that’s better (and much, much faster) than you were going to do on your own anyway. LLMs are a game-changer for trying and learning new languages, skills, tools... the list goes on.
I could be biased here. I'm a small business owner, and I wear a ton of hats, so I struggle with learning something new almost every day. But here's a Reddit thread with people describing the positive impact of ChatGPT:
“A co-worker of mine often mentions how things were in Brazil so I am using the platform to learn some conversational Portuguese.”
“I am interested in theoretical physics but am not a scientist. I was just asking it to help me understand a topic I don't know but heard about on a podcast.”
“I've used it to write some job descriptions for people I need to hire into my team.”
“I've used it to walk me through the pros and cons of switching from an electric to an induction stove.”
“I've been using it to find new productivity apps to work more efficiently. It told me about bitwarden, Notion, and Figma. Then it taught me how to use those apps.”
“I had a rudimentary understanding of Lua a couple weeks ago. ChatGPT has helped me learn it way better than any video tutorials or websites has ever helped me learn a programming language.”
Untangling the hype
There’s another technology concept called Amara’s law. In short, it states:
We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.
Based on the linked Reddit thread, it’d be easy to look at the use cases cited and think of them as small fry. But there are a few things to consider.
Right now, we’re in the earliest stages of ChatGPT (which is why it’s easy to get carried away with the hype). Realistically though, the software will only improve from here. This is only the fourth version of OpenAI’s GPT model (and OpenAI has only been around for eight years). What does version 5, version 10, or version 100 look like? OpenAI may never completely solve the hallucination problem, but they can probably get most of the way there.
Yet, even as the underlying model improves, there are limits to what a simple chatbot UI can do. Tailoring a product for a specific audience creates much more value, but takes time. Imagine going from “learn some conversational Portuguese” to “have a full-time Portuguese tutor and speaking partner in my pocket.” Companies that can deeply understand their customers and build better UX on top of GPT models will win big5.
Sam Altman, the CEO of OpenAI, has called ChatGPT a “horrible product,” given how often it breaks. But there are already non-technical, non-early adopters who are using ChatGPT for their day-to-day workflows. Companies are using AI to write sales emails and marketing content. Authors are writing books with GPT3, then using Stable Diffusion to create the cover art. But this many use cases, this quickly, indicates that ChatGPT (or its successors) will stick around for a while to come.
If you’re a reader of Matt Levine, you’ll remember when Long Island Iced Tea rebranded to Long Blockchain and its stock jumped 200%.
This, by the way, was the promise of Siri and Alexa several years ago that never quite panned out. I’m interested to see how the big tech players use LLMs to revamp their voice assistants in the coming months.
Instagram has really hammered home this point for me. Whether it was Snapchat’s Stories or TikTok’s infinite video feed, Instagram has demonstrated that you can get away with quickly copying the best features of your competitors. As long as your user base is big enough, most of your users are happy to stay put.
I am deliberately avoiding the ongoing debate about AI-generated content and copyright. It’s a much longer, nuanced discussion that I want to explore, and both sides of the argument have merit. Technology is always a double-edged sword, for better and worse. For now, let’s assume that ChatGPT is having a net positive impact.
And there are infinite customer niches! I can already see parallels between AI and the history of internet content or the history of online shopping. If you can find a reasonably small customer segment, you can dial in your marketing and customer experience well enough to capture that niche.